I have a continuous bipolar signal that can vary between ±1 mV and I would like to figure out the best way to measure it with high resolution without introducing a lot of noise.

My first thought is to amplify the signal with a instrument amplifier (looking at INA333) and digitize with a MCP3421 18-bit differential ADC. Using a gain of 8 and internal 2.048 V ref on the MCP3421, I believe the FSR is ±256 mV. I'd like to amplify my signal to use this FSR and get as much resolution as I can.

  1. Is using an instrument amplifier the best way to go about this?

  2. How can I use an instrument amp and maintain a bipolar output to feed to the ADC (do I use 2 opposed instrument amps?)

  3. Is this possible to do without using a negative power supply for the reference? My board will be 0-5 V.

  4. Any suggestions for the best way to do this are welcome (and greatly appreciated). I would love to see some schematics too if possible!

  • 1
    \$\begingroup\$ What is the spectrum of interest of the signal? \$\endgroup\$
    – Andy aka
    Aug 2, 2016 at 17:49
  • \$\begingroup\$ @Andyaka Hi Andy, this is a signal from a differential thermocouple. Im trying to measure the small voltage difference between the pair's ends as I vary some other parameter. Basically a DC signal that changes very slowly. \$\endgroup\$
    – SudoKill
    Aug 2, 2016 at 18:32
  • 1
    \$\begingroup\$ "As much resolution as you can" isn't a goal worth reaching for. Sit down and figure out how much resolution you need, then compare that to how much resolution you can get. For example, with a gain of 200 and a 2V reference, do you really need better than 1/4096? \$\endgroup\$ Aug 3, 2016 at 1:08

2 Answers 2


Is using an instrument amplifier the best way to go about this?

Yes it is unless you are going to use an isolated combined ADC and InAmp (as purveyed by Analog Devices and very nice they are too). I'll try and remember the part number if pressed.

Given that a thermocouple will likely be deliberately or inadvertently shorted to ground/earth I have to suggest using an InAmp with a split rail supply because the polarity may not be known and the signal voltage may indeed go negative to ground. Also it keeps open options for best common mode noise rejection. You can make a negative supply from a positive one - there are chips that do this and quite low power ones too.

Then, what you do is offset the reference pin on the InAmp (pin 5) to the middle on your ADC range so that you can take a full measurement if polarity is switched. Top half of the ADC range deals with correctly connected thermocouples and the bottom half deals with incorrectly connected thermocouples.

Given also that the input offset of your InAmp is roundabout equivalent to 1 degC you may choose to multiplex (using a DG409 for instance) so you can auto zero. This then gives you the possibility of also measuring a cold junction reference and maybe a precision voltage reference to give you some accurate gain calibration point.

To what extent you go down this line is up to you.


Use an inamp like the AD623, with input ranges that can go below the rails. Use a 2.5v reference to supply a bias to the Vref pin on the inamp, and use a single resistor to set the gain to about 200 or more, if there really is no DC bias on the input. Use a 12 or 16 bit ADC with the reference set to about a volt, and you'll have plenty of resolution.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.